See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/8535851

An Improved Method to Analyze the Stress Relaxation of Ligaments Following a

Finite Ramp Time Based on the QuasiLinear Viscoelastic Theory

Article in Journal of Biomechanical Engineering · February 2004

DOI: 10.1115/1.1645528 · Source: PubMed

CITATIONS READS

128 584

2 authors:

Steven D Abramowitch Savio L-Y. Woo

University of Pittsburgh University of Pittsburgh
102 PUBLICATIONS 2,891 CITATIONS 526 PUBLICATIONS 31,748 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Steven D Abramowitch on 20 May 2014.

The user has requested enhancement of the downloaded file.


Steven D. Abramowitch
Savio L.-Y. Woo*
Musculoskeletal Research Center
Department of Orthopaedic Surgery
Department of Bioengineering
University of Pittsburgh
E1641 Biomedical Science Tower
210 Lothrop Street
P.O. BOX 71199
Pittsburgh, PA 15213
Phone: 412-648-2000
FAX: 412-648-2001
e-mail: ddecenzo@pitt.edu
An Improved Method to Analyze
the Stress Relaxation of
Ligaments Following a Finite
Ramp Time Based on the
Quasi-Linear Viscoelastic Theory
The quasi-linear viscoelastic (QLV) theory proposed by Fung (1972) has been frequently
used to model the nonlinear time- and history-dependent viscoelastic behavior of many
soft tissues. It is common to use five constants to describe the instantaneous elastic
response (constants A and B) and reduced relaxation function (constants C, ␶ 1 , and ␶ 2 兲
on experiments with finite ramp times followed by stress relaxation to equilibrium. How-
ever, a limitation is that the theory is based on a step change in strain which is not
possible to perform experimentally. Accounting for this limitation may result in regression
algorithms that converge poorly and yield nonunique solutions with highly variable con-
stants, especially for long ramp times (Kwan et al. 1993). The goal of the present study
was to introduce an improved approach to obtain the constants for QLV theory that
converges to a unique solution with minimal variability. Six goat femur-medial collateral
ligament-tibia complexes were subjected to a uniaxial tension test (ramp time of 18.4 s)
followed by one hour of stress relaxation. The convoluted QLV constitutive equation was
simultaneously curve-fit to the ramping and relaxation portions of the data 共 r 2 ⬎0.99兲 .
Confidence intervals of the constants were generated from a bootstrapping analysis and
revealed that constants were distributed within 1% of their median values. For validation,
the determined constants were used to predict peak stresses from a separate cyclic stress
relaxation test with averaged errors across all specimens measuring less than 6.3⫾6.0%
of the experimental values. For comparison, an analysis that assumed an instantaneous
ramp time was also performed and the constants obtained for the two approaches were
compared. Significant differences were observed for constants B, C, ␶ 1 , and ␶ 2 , with ␶ 1
differing by an order of magnitude. By taking into account the ramping phase of the
experiment, the approach allows for viscoelastic properties to be determined independent
of the strain rate applied. Thus, the results obtained from different laboratories and from
different tissues may be compared. 关DOI: 10.1115/1.1645528兴

Introduction iterative techniques, as well as extrapolation and deconvolution

The nonlinear time- and history-dependent viscoelastic behav-
ior of soft biological tissues has been widely described by the
quasi-linear viscoelastic 共QLV兲 theory formulated by Fung 共1972兲
关1–12兴. This theory has been adopted for modeling the viscoelas-
tic behavior of ligaments and tendons by many laboratories
关9,13–17兴. In our laboratory, the QLV theory has been success-
fully used for modeling the viscoelastic properties of the canine
medial collateral ligament 共MCL兲 关18兴, the porcine anterior cruci-
ate ligament 关19兴, and the human patellar tendon 关20兴.
In the QLV theory, the reduced relaxation function, with con-
stants C, ␶ 1 , and ␶ 2 , describes the time-dependent stress relax-
ation of a tissue normalized by the stress at the time of a step input
of strain. Since it is experimentally impossible to apply a step
increase in strain, many previous investigators have applied ex-
tensions at relatively high rates. Nevertheless, this approach
proves to be technically challenging as it is difficult to measure
strain accurately. On the other hand, others have modified the
analysis to account for finite ramp times in order to better approxi-
mate solutions. These methods include normalization procedures,
*Corresponding author.
Contributed by the Bioengineering Division for publication in the JOURNAL OF
BIOMECHANICAL ENGINEERING. Manuscript received by the Bioengineering Divi-
sion April 2, 2003; revision received October 2, 2003. Associate Editor:
L. Soslowsky.
关2,7,19,21,22兴. However, these methods are still dependent on fast
ramp times 共⬃0.01 to 0.1 s兲 and therefore may be affected by the
errors associated with high extension rates 共eg., overshoot, vibra-
tion, poorly approximated strain histories兲 关14兴. Previously, a two-
step regression approach was used by our laboratory to improve
the mathematical representation of the experimentally obtained
reduced relaxation function 关19兴. Because two physically signifi-
cant constants 共B and ␶ 1 ) become highly correlated with this ap-
proach, it should be recognized that regression algorithms may
fail to converge for certain applications. Further, the obtained so-
lutions may be non-unique and constants may be highly sensitive
to systematic deviations between the model and experimental data
as well as random noise.
Therefore, the objective of this study was to develop an im-
proved analytical method to obtain the constants of QLV theory
共A, B, C, ␶ 1 , and ␶ 2 ). By simultaneously curve-fitting the QLV
constitutive equation 共i.e., convolution integral of the instanta-
neous elastic response and reduced relaxation function兲 to the
ramping and relaxation portions of the data from a static stress-
relaxation test, algorithms converge to unique solutions with mini-
mal variability of the constants for long ramp times. Therefore,
the actual strain history can be accurately approximated and errors
resulting from fast extension rates can be avoided. Six goat femur-
medial collateral ligament-tibia complexes were subjected to a
uniaxial tension test 共ramp time of 18.4 s兲 followed by one hour of

92 Õ Vol. 126, FEBRUARY 2004 Copyright © 2004 by ASME Transactions of the ASME
stress relaxation. A bootstrapping analysis was performed to as-
sess the sensitivity of the constants to systematic deviations be-
tween the model and experimental data, random noise, and nu-
␴共 t 兲⫽ 冕 t

⫺⬁
G 共 t⫺ ␶ 兲
⳵␴ e 共 ␧ 兲 ⳵ ␧
⳵␧ ⳵␶
⳵␶ (2)

merical instabilities. For validation, the constants obtained were
used to predict the results of a separate cyclic stress relaxation
experiment. For comparison, an analysis that assumed an instan-
taneous ramp time was also performed and the constants obtained
for the two approaches were compared.
Materials and Methods
Experimental Studies. Six femur-MCL-tibia complexes
共FMTCs兲 were obtained from six Sannen breed goats 共wt. 38.0
⫾4.7 kg; mean⫾sd兲. These specimens were contra-lateral sham-
operated controls used in a separate study of the structure and
function of the healing MCL in response to a combined ACL/
MCL injury 关23兴. As the focus of this paper is to present an ana-
lytical approach, please refer to the referenced articles for details
regarding experimental methodology including specimen prepara-
tion, cross-sectional area measurement, strain tracking, and set-up
for mechanical testing 关24 –28兴. As described in our previous
study, the cross-sectional area of the MCL was measured to be
8.6⫾1.7 mm2 共mean⫾SD兲 关23兴. Our experimental protocol is also
In the experimental setting, we can assume that the history
begins at t⫽0. For soft tissues whose ␴-␧ relationship and hyster-
esis are not overly sensitive to strain rate, Fung has proposed the
following expression for G(t) based upon a continuous spectrum
of relaxation:
1⫹C 关 E 1 共 t/ ␶ 2 兲 ⫺E 1 共 t/ ␶ 1 兲兴
G共 t 兲⫽ (3)
1⫹C ln共 ␶ 2 / ␶ 1 兲
where E 1 (y)⫽ 兰 ⬁y e ⫺z /zdz is the exponential integral, and C, ␶ 1 ,
and ␶ 2 are material constants. An exponential approximation has
been chosen to describe the instantaneous elastic response:
␴ e 共 ␧ 兲 ⫽A 共 e B␧ ⫺1 兲 (4)
where A and B are material constants 关29兴.
Approaches for Constant Estimation. For the current ap-
proach 共which will be termed ‘‘strain history approach’’ through-
out the remainder of this paper兲, the stress resulting from a ramp
phase with a constant strain rate ␥ over the times 0⬍t⬍t 0 can be
written by substituting Eqs. 共3兲 and 共4兲 into Eq. 共2兲:

冕
described previously 关23兴. Briefly, each specimen was attached to AB ␥ t
a materials testing machine 共Instron™兲 within a saline bath that ␴ 共 t:0⬍t⬍t 0 , ␪ 兲 ⫽ 兵 1⫹C共 E 1 关共 t⫺ ␶ 兲 / ␶ 2 兴
was kept at a constant 37°C. Each specimen was preloaded to 2 N 1⫹C ln共 ␶ 2 / ␶ 1 兲 0
and the gauge length was zeroed. This was followed by precon-
ditioning between 0 and 1.5 mm of extension. Prior to each test, ⫺E 1 关共 t⫺ ␶ 兲 / ␶ 1 兴 兲 其 e B␥␶ ⳵ ␶ (5)
specimens were held at a zero-load position and allowed to equili- where ␪ ⫽ 兵 A,B,C, ␶ 1 , ␶ 2 其 .
brate within the saline bath for one hour. Similarly, the subsequent stress relaxation from t 0 to t⫽⬁, can
Each FMTC then underwent a static stress relaxation test be described as

冕
whereby specimens were elongated to 3 mm at 10 mm/min and
held for a period of 60 min 关23兴. Preliminary tests revealed that 3 AB ␥ t0
␴ 共 t:t⭓t0 , ␪ 兲 ⫽ 兵 1⫹C共 E 1 关共 t⫺ ␶ 兲 / ␶ 2 兴
mm of elongation resulted in midsubstance strains less than 5%. 1⫹C ln共 ␶ 2 / ␶ 1 兲 0
An elongation rate of 10 mm/min was chosen to represent a typi-
cal rate for tensile testing protocols. Thus, the time until peak load ⫺E 1 关共 t⫺ ␶ 兲 / ␶ 1 兴 兲 其 e B␥␶ ⳵ ␶ (6)
was t0 ⫽18.4 s. Strain increased linearly with time, thus this elon-
gation rate resulted in an average midsubstance strain rate, ␥, of For a set of experimental data, the ramping portion of the data
0.15%/s. The advantage of utilizing a slow strain rate is that the was defined as (t i ,Ri), from 0⬍t i ⬍t 0 and the relaxation data as
actual strain history can be well approximated by a linear ramp (t i ,Si), from t 0 to t⫽⬁. Thus, the sums of squares difference
followed by a hold at a constant strain magnitude. Thus, the errors between the experimentally obtained data and the theory can be
associated with fast strain rates are avoided 共eg., overshoot, vibra- expressed as:
tion, poorly approximated strain histories兲 关14兴. For these data, the
total percentage of stress relaxation was defined as the difference f 共 ␪ 兲⫽ 兺 关 R ⫺ ␴ 共 t :0⬍t ⬍t
i
i i i 0 , ␪ 兲兴
2
(7)
between the peak stress at t0 and the stress measured at the end of
the test, normalized by the peak stress. The nonlinear stress-strain and
curve was determined from the ramping phase of this test.
Following one hour of recovery, a second test was performed to
measure the cyclic stress relaxation behavior of the MCL. In this g共 ␪ 兲⫽ 兺 关 S ⫺ ␴ 共 t :t ⭐t ⬍t, ␪ 兲兴
i
i i 0 i
2
(8)
test, each FMTC was subjected to eight cycles of elongation be-
tween 2 and 3 mm at 10 mm/min and the corresponding peak and Since Eqs. 共7兲 and 共8兲 are both functions of ␪, the strain history
valley stresses were recorded. This corresponded to physiologic approach minimizes these equations simultaneously using a non-
strains of approximately 1 to 3.5%, respectively. Data was col- linear optimization algorithm. The algorithm used in this study
lected at a constant rate of 5 Hz throughout all tests. was a Levenberg-Marquardt algorithm that was modified to mini-
mize f共␪兲⫹g共␪兲. The code for this algorithm was written using
Quasilinear Viscoelastic Theory. The QLV theory assumes Mathematica 共Wolfram Research, Inc. Champaign, IL兲, and was
that the stress relaxation behavior of soft-tissue can be expressed largely based on the algorithm outlined in the book Numerical
as: Recipes in C 关30兴.
␴ 共 t 兲 ⫽G 共 t 兲 * ␴ c共 ␧ 兲 (1) This strain history approach gives a direct fit of Eqs. 共7兲 and 共8兲
to the experimental data with no modification of the theory or
where ␴ (␧) is the instantaneous elastic response, i.e., the maxi-
c
normalization of the data. Assuming that the theory provides a
mum stress in response to an instantaneous step input of strain, ␧. perfect Gaussian curve-fit of the data and a global minimum of
G(t) is the reduced relaxation function that represents the time- f共␪兲⫹g共␪兲 can be uniquely determined, the constants obtained that
dependent stress response of the tissue normalized by the stress at describe the instantaneous elastic response 共A and B兲 and the
the time of the step input of strain 关i.e., t⫽0 ⫹ , such that G(t) spectrum of relaxation 共C, ␶ 1 , and ␶ 2 ) would be those obtained if
⫽ ␴ (t)/ ␴ (0 ⫹ ), and G(0 ⫹ )⫽1]. a true step-elongation were to have been applied. This is because
Using the Boltzmann superposition principle, the stress at time the actual strain history can be well approximated at slow exten-
t, ␴ (t), is given by the convolution integral of the strain history sion rates. However, further analysis of the strain history approach
and G(t): revealed that curve-fits were non-Gaussian and constants A and ␶ 1

Journal of Biomechanical Engineering FEBRUARY 2004, Vol. 126 Õ 93


Fig. 1 A typical residual plot demonstrating systematic devia-
tions of the model and experimental data
became significantly correlated as ramp time increases. Since con-
stant A is not required to describe the reduced relaxation following
ramping at a constant strain rate 关19兴, it was held fixed for each
individual regression analysis. Its value was determined by curve-
fitting Eq. 共4兲 to the ramping portion of the experimental data.
This is further justified by previous work which has shown that
the stress-strain curves of ligaments are insensitive to strain rates
ranging over four decades 关31兴.
The initial guess was determined from a preliminary analysis.
To ensure global convergence, the initial guess for each constant
was multiplied by a random factor between 0.1 and 10 and the
regression analysis was performed. This was done 100 times and
the guess that produced the lowest minimum of the objective
function was the guess chosen for all subsequent analyses. It was
found that the algorithm was relatively insensitive to the initial
guess; that is, the algorithm consistently converged to a unique
solution which was assumed to be the global minimum.
Because the curve-fits were non-Gaussian, it was necessary to
assess the variability of constants that may result from systematic
deviations between the model and the experimental data, experi-
mental noise, and numerical instabilities. This study used a boot-
strapping analysis that was previously described by Yin et al. 关32兴.
First, the data from each specimen was curve-fit individually us-
ing the theory. Residual plots for each specimen were then curve-
fit with a polynomial function to obtain a pool of curves repre-
senting systematic error 共Fig. 1兲. Taking the difference between
the systematic error curves and the residuals allowed for a random
noise distribution for each specimen to be obtained 共Fig. 2兲. Fi-
nally, a randomly selected systematic error curve and random er-
ror distribution were added to predicted stresses that were ob-
Fig. 2 A typical random error plot with distribution 0Á0.00874
„meanÁSD…
94 Õ Vol. 126, FEBRUARY 2004
tained by curve-fitting the theory to averaged experimental data.
This created a set of data whose systematic error and random
noise distributions were representative of those that could be ob-
served experimentally. Constants A, B, C, ␶ 1 , and ␶ 2 for this new
data set were obtained using the strain history approach. One-
hundred new data sets were generated using this methodology and
95% confidence intervals for each constant were obtained.
If the variability of a constant with respect to its median value
共i.e., the difference between upper bound of the interval and the
median, normalized by the median兲 was large 共eg., ⬎5%兲, then
that constant is sensitive to systematic deviations between the
model and the experimental data, experimental noise, and numeri-
cal instabilities. Thus, there would be less confidence that the
solutions obtained from an individual regression analysis were
meaningful since a data set with a slightly different distribution of
random noise or systematic error may result in a completely dif-
ferent solution.
For validation of the constants, the ability of QLV theory to
predict the results of the cyclic stress relaxation test using the
constants obtained from the 60-min stress relaxation test was as-
sessed. The constants A, B, C, ␶ 1 , and ␶ 2 describing each speci-
men were separately substituted into Eq. 共2兲 to obtain the specific
equation describing the stress response of each goat FMTC. Using
a previously described approximated cyclic strain-time history
based 关18兴, the peak stresses obtained from Eq. 共2兲 were then
compared to those obtained experimentally during the cyclic
stress relaxation test for each specimen.
For comparison, the strain history approach was compared to
another approach to estimate the five constants of Eqs. 共3兲 and 共4兲.
The instantaneous assumption approach as described by Woo
et al. 共1981兲 assumes that the ramping phase of the stress relax-
ation tests occurred instantaneously 关18兴. Therefore, G共t兲 and
␴ c(␧) are assumed to be separable functions. Constants can be
obtained by fitting Eq. 共4兲 to (␧ i ,Ri), where ␧ i⫽ ␥ t i is the experi-
mental strain from 0⬍t⬍t 0 , and fitting Eq. 共5兲 to (t i ,si), where si
is equal to Si normalized by the stress at t 0 . Thus, this approach
does not account for relaxation that occurred during ramping
which may result in increasingly erred estimates of constants as
ramp times increase 关33兴.
As the results from the bootstrapping analysis showed relatively
small variations in the constants between the two approaches, so-
lutions obtained for each individual regression analysis were con-
sidered to be stable. Thus, solutions obtained from curve-fitting
each individual specimen’s data with the two approaches could be
compared. A nonparameteric two-tailed Wilcoxon sign-rank test
and z-score statistic were used for statistical comparison and to
calculate p-values. Significance was set at p⬍0.05.
Results
During the ramping phase of the test, the stress-strain curves
exhibited linear behavior up to approximately 1.5% strain. After-
ward, the curve became nonlinear with the stress at the end of the
ramping phase measuring 15.3⫾5.4 MPa. The stress-relaxation
phase was nonlinear with respect to time as the largest percentage
of relaxation occurred in the first 10 min. After, 40 min the rate of
relaxation had diminished, with only a 0.6⫾0.4% reduction over
the remaining 20 min of the test. At 60 min, the total amount of
stress relaxation was 31.9⫾11.5%.
A typical curve-fit to the ramping and stress relaxation portions
of the experimental data using the strain history approach is de-
tailed in Fig. 3. These data were fit with an overall r2 value greater
than 0.99. Similarly, the instantaneous assumption approach was
able to achieve separate fits of these data with good quality (r2
⬎0.99 for both portions兲. In terms of the constants describing the
instantaneous elastic response, the confidence intervals for con-
stant A obtained from the boot-strapping analysis were 5.8:6.0
MPa 共lower bound 95% CI:upper bound 95% CI兲 for both ap-
proaches as they both determine constant A in the same manner.
All values for constant A were within 1.7% of its median value.
Transactions of the ASME
Fig. 3 A typical curve fit using the strain history approach to
experimental data „␥Ä0.15%Õs during ramping…
Fig. 4 The reduced relaxation function as determined using
the instantaneous assumption approach and the strain history
approach

Table 1 Constants describing the instantaneous elastic re- sults of those analyses are presented in Tables 1 and 2. It should
sponse obtained by curve-fitting individual specimens using
be noted that these were paired comparisons and the high standard
the strain history and instantaneous assumption approaches.
* significant difference between the two approaches „pË0.05…. deviations are the result of inter-specimen variability. For these
data it is important to consider whether a particular constant was
A 共MPa兲 B determined to be either consistently higher or lower for specimens
when determined using the strain history approach 共Tables 1 and
Specimen # Strain History Inst. Assum. Strain History Inst. Assum.
2兲.
997 6.75 6.75 73.5 61.8 In terms of the solutions obtained from curve-fits of each indi-
965 3.04 3.04 78.5 76.3 vidual specimen’s data, constant B was determined to be signifi-
422 2.75 2.75 76.8 75.5
41 7.19 7.19 41.6 38.5 cantly greater when obtained using the strain history approach
26 32.86 32.86 13.9 13.5 共53.1⫾27.0; mean⫾SD兲 compared to the instantaneous assump-
19 5.65 5.65 34.0 33.1 tion approach 共49.8⫾25.4兲, indicating a more nonlinear instanta-
Mean⫾SD 9.7⫾11.5 9.7⫾11.5 53.1⫾27.0* 49.8⫾25.4 neous elastic response 共p⬍0.05; Table 1兲. For the constants de-
scribing the reduced relaxation function, the strain history
approach consistently estimated significantly higher values for
constant C 共0.089⫾0.057 versus 0.076⫾0.053兲 for each indi-
For constant B, the confidence intervals for the strain history ap- vidual specimen 共p⬍0.05兲. Estimates for constant ␶ 1 obtained us-
proach and instantaneous assumption approach were 47.2:48.2 ing the strain history approach 共0.54⫾0.15 s兲 were an order of
versus 45.7:46.4, respectively. For each approach, all constants magnitude lower than those obtained using the instantaneous as-
were determined within approximately 1% of their median values. sumption approach 共2.13⫾0.98 s; p⬍0.05兲. Further, estimates for
In terms of the constants describing the reduced relaxation func- constant ␶ 2 were also significantly lower when determined with
tion, the bootstrapping analysis yielded confidence intervals for the strain history approach 共1602⫾581 s versus 2222⫾821 s;
constant C from the strain history approach and the instantaneous p⬍0.05兲. Thus, it can be seen that the strain history approach
assumption approach to be 0.0721:0.0724 versus 0.0680:0.0680, consistently predicts a reduced relaxation function with a greater
respectively. Thus, all values determined for this constant were percentage of relaxation, steeper initial slope, and earlier time to
within 1% of the median values for both approaches. Similarly, reach equilibrium 共Fig. 4兲.
the confidence intervals for constants ␶ 1 共0.62 s:0.63 s versus 2.01 For validation, the constants A, B, C, ␶ 1 , and ␶ 2 obtained from
s:2.02 s兲 and ␶ 2 共1469 s:1488 s versus 2138 s:2145 s兲 for the strain history approach could accurately describe the experi-
the strain history versus instantaneous assumption approaches mental data of the cyclic stress relaxation test for each specimen.
also determined values to be within 1% of their median values, Error between the prediction and experimental data ranged from
respectively. 0.2% to 2.9% for the best prediction 共Fig. 5a兲 and 9.3% to 16.2%
As the boot-strapping analysis demonstrated minimal sensitiv- for the worst prediction 共Fig. 5b兲. In general, the prediction of the
ity of the obtained constants, it was deemed that the solutions to initial peak stress was the most erred for all specimens. Nonethe-
both approaches are stable. Thus, the constants obtained for each less, the average error for this peak measured only 6.3⫾6.0%
individual specimen’s data can be compared statistically. The re- across all specimens.

Table 2 Constants describing the reduced relaxation function obtained by curve-fitting indi-
vidual specimens using the strain history and instantaneous assumption approaches. * sig-
nificant difference between the two approaches „pË0.05….

C ␶ 1 (sec… ␶ 2 (sec…
Specimen # Strain History Inst. Assum. Strain History Inst. Assum. Strain History Inst. Assum.
997 0.204 0.183 0.30 3.29 1972 2451
965 0.055 0.047 0.64 2.07 1997 2685
422 0.071 0.058 0.63 1.85 896 1163
41 0.084 0.073 0.54 2.97 2248 3030
26 0.066 0.050 0.44 0.51 1587 2786
19 0.056 0.045 0.70 2.09 910 1218
Mean⫾SD 0.089⫾0.057* 0.076⫾0.053 0.54⫾0.15* 2.13⫾0.98 1602⫾581* 2222⫾821

Journal of Biomechanical Engineering FEBRUARY 2004, Vol. 126 Õ 95


Fig. 5 Prediction of the peak stresses of a cyclic loading his-
tory based on the constants obtained from the stress relax-
ation experiment using the strain history approach for indi-
vidual specimens „a… best prediction; „b… worst prediction
Discussion
In this study, an improved analytical method to obtain the con-
stants of Fung’s QLV theory 共A, B, C, ␶ 1 , and ␶ 2 ) was presented.
The current approach takes into account the ramping portion of
the experiment; therefore, removing the assumption of a step
change in strain. The strain history approach enabled us to fit the
entire experiment 共i.e., the stress versus time data from the ramp-
ing phase to the stress relaxation portion of the stress relaxation
test兲. The key finding was that the regression algorithm converged
to a unique solution that was stable and validated by predicting
the stress response of a separate experiment. By comparing the set
of constants for individual specimens to those obtained for the
instantaneous assumption approach, the effect of the assumption
of a step change was elucidated.
Previous studies using QLV theory together with the assump-
tion of a step change in strain do not account for the relaxation
that occurred during ramping and base normalization of experi-
mental data on an underestimated peak stress 关11,33,34兴. As a
result, constants C and ␶ 1 are significantly underestimated and
overestimated, respectively 关33兴. The current approach improves
the estimates of these constants. The constant C was significantly
higher and constant ␶ 1 was an order of magnitude lower than
those determined using the instantaneous assumption approach.
Further, the strain history approach estimates a more nonlinear
instantaneous elastic response as indicated by the higher value for
constant B. All of these improved approximations result from ac-
counting for the strain history during ramping.
In spite of the fact that the ramp time used in this study is 1–2
orders of magnitude longer than those used previously, the con-
stants obtained using the strain history approach are similar to
those reported previously for ligaments and tendons
关17,18,20,35兴. Most importantly, the determination of constant ␶ 1 ,
which has been shown to be the most sensitive to the experimental
ramp time, was on the same order of magnitude as these previous
studies 关33兴. Further, the ability for the model to predict the results
of a second experiment for validation of the constants was similar
to that reported by Woo et al., 1981 whose constants were also
determined based on data with a ramp time that was 2 orders of
magnitude shorter than that used in this study. While this theory is
only an approximation and no approach can guarantee that the
obtained constants are ‘‘true,’’ the fact that the strain history ap-
proach was able to estimate reasonable constants based on data
with a ramp time 1–2 orders of magnitude slower than previous
96 Õ Vol. 126, FEBRUARY 2004
Table 3 Constants describing the instantaneous elastic re-
sponse and reduced relaxation function obtained by curve-
fitting individual specimens using the approach described by
Kwan et al. „1993…. ␺ denotes convergence failure of the algo-
rithm, for which the constants obtained for the iteration prior to
failure are reported.
Specimen # A 共MPa兲 B C ␶ 1 (sec… ␶ 2 (sec…
997␺ ⬍1E⫺10 2.1E⫹07 0.173 4.87 2287
␺
965 15.62 36.3 0.068 5.25E⫺06 1275
422 10.95 46.2 0.071 9.05E⫺05 1568
41␺ ⬍1E⫺10 2.5E⫹07 0.072 4.16 3186
26 1.68 90.0 0.071 1.34E⫺04 1657
19 2.67E⫺10 548.5 0.046 1.52 1197
studies is a significant advancement. Thus, issues associated with
fast strain rates can be avoided 共eg., overshoot, vibration, poorly
approximated strain histories兲 关14兴.
Previously our laboratory had used a two-step regression ap-
proach to obtain better estimates of the constants 共A, B, C, ␶ 1 ,
and ␶ 2 ) based on experiments with finite ramp times 关19兴. For this
approach, however, two physically significant constants, B and
␶ 1 , are highly correlated. As a result, regression algorithms either
fail to converge or converge to solutions that are sensitive to sys-
tematic deviations between the theory and the experimental data
for certain applications. This is demonstrated in Table 3 where the
constants 共A, B, C, ␶ 1 , and ␶ 2 ) were determined for each indi-
vidual specimen using this approach. Convergence failed for three
of six specimens and the constants A, B, and ␶ 1 spanned more
than 10 orders of magnitude.
A similar issue 共i.e., the correlation of constants A and ␶ 1 ) was
observed using the strain history approach. It was found that a
relatively similar curve-fit could be achieved when both constants
A and ␶ 1 are increased and decreased by the same percentage of
their original value, respectively. Thus, this study chose to fix
constant A based on work from two previous studies. One study
showed that constant A is unnecessary to describe the reduced
stress relaxation following a finite ramp time, and the other dem-
onstrated that the stress-strain curve of ligaments is relatively in-
sensitive to strain-rates ranging over 4 decades 关19,31兴. This al-
lowed for a significant improvement of the variability displayed
for the obtained constants when compared to that of Kwan et al.
共1993兲. It should be noted that this study did not compare to other
approaches that account for the loading portion of the experiment
as these approaches are fundamentally based on short ramp times
关2,7,22兴. These approaches were not developed to apply to the
experimental protocol utilized in this study, but will perform ad-
equately when applied appropriately.
For tissues whose stress-strain curve is sensitive to strain-rate,
an alternative means to determine constant A may be necessary.
For this case, a high strain-rate experiment may still be required to
determine constant A, but the remaining constants could be deter-
mined utilizing a second slower strain rate experiment along with
the approach described in this study. Although this methodology
would still require accurate high-speed measurements of stress
and strain to determine constant A, the errors caused by the in-
ability of a testing machine to accurately perform a ramp and hold
test at high strain rates can still be avoided.
Because of the stability of the solutions obtained using the
strain history approach, constants obtained from curve-fitting in-
dividual specimen’s data can be compared. Thus, the strain history
approach may prove to be an important analytical tool when mak-
ing statistical comparisons of the viscoelastic properties obtained
from different laboratories or from different tissues. In addition,
researchers can focus on collecting accurate measurements of
stress and strain at slower strain rates while still having the ability
to determine meaningful constants describing the viscoelastic be-
havior of their specimens. We believe the strain history approach
may allow for differences in the constants to be determined that
Transactions of the ASME
 may have otherwise been unobserved due to the effects of a finite
ramp time or performing experiments at fast strain rates. This may
eventually allow for more accurate comparisons of the constants
describing injured, diseased, or treated tissues versus those of nor-
mal tissues so that researchers can quantitatively measure differ-
ences in viscoelastic behavior. In the future, we hope to use this
approach to determine the differences in the viscoelastic behaviors
of healing and sham operated MCLs.
Acknowledgment
The authors acknowledge the financial support provided by the
National Institute of Health Grant AR41820.
References
关1兴 Fung, Y. C., 1972, ‘‘Stress Strain History Relations of Soft Tissues in Simple
Elongation,’’ in Biomechanics: Its Foundations and Objectives, eds., Y. C.
Fung, N. Perrone, and M. Anliker, PrenticeHall, Englewood Cliffs, NJ, pp.
181–207.
关2兴 Carew, E. O., Talman, E. A., Boughner, D. R., and Vesely, I., 1999, ‘‘Quasi-
Linear Viscoelastic Theory Applied to Internal Shearing of Porcine Aortic
Valve Leaflets,’’ J. Biomech. Eng., 121共4兲, pp. 386 –392.
关3兴 Huang, C. Y., Mow, V. C., and Ateshian, G. A., 2001, ‘‘The Role of Flow-
Independent Viscoelasticity in the Biphasic Tensile and Compressive Re-
sponses of Articular Cartilage,’’ J. Biomech. Eng., 123共5兲, pp. 410– 417.
关4兴 Huyghe, J. M., van Campen, D. H., Arts, T., and Heethaar, R. M., 1991, ‘‘The
Constitutive Behavior of Passive Heart Muscle Tissue: A Quasi-Linear Vis-
coelastic Formulation,’’ J. Biomech., 24共9兲, pp. 841– 849.
关5兴 Iatridis, J. C., Setton, L. A., Weidenbaum, M., and Mow, V. C., 1997, ‘‘The
Viscoelastic Behavior of the Non-Degenerate Human Lumbar Nucleus Pulpo-
sus in Shear,’’ J. Biomech., 30共10兲, pp. 1005–1013.
关6兴 Kim, S. M., McCulloch, T. M., and Rim, K., 1999, ‘‘Comparison of Viscoelas-
tic Properties of the Pharyngeal Tissue: Human and Canine,’’ Dysphagia,
14共1兲, pp. 8 –16.
关7兴 Myers, B. S., McElhaney, J. H., and Doherty, B. J., 1991, ‘‘The Viscoelastic
Responses of the Human Cervical Spine in Torsion: Experimental Limitations
of Quasi-Linear Theory, and a Method for Reducing These Effects,’’ J. Bio-
mech., 24共9兲, pp. 811– 817.
关8兴 Provenzano, P., Lakes, R., Keenan, T., and Vanderby, Jr., R., 2001, ‘‘Nonlinear
Ligament Viscoelasticity,’’ Ann. Biomed. Eng., 29共10兲, pp. 908 –914.
关9兴 Puso, M. A., and Weiss, J. A., 1998, ‘‘Finite Element Implementation of An-
isotropic Quasi-Linear Viscoelasticity Using a Discrete Spectrum Approxima-
tion,’’ J. Biomech. Eng., 120共1兲, pp. 62–70.
关10兴 Rousseau, E. P., Sauren, A. A., van Hout, M. C., and van Steenhoven, A. A.,
1983, ‘‘Elastic and Viscoelastic Material Behavior of Fresh and
Glutaraldehyde-Treated Porcine Aortic Valve Tissue,’’ J. Biomech., 16共5兲, pp.
339–348.
关11兴 Sauren, A. A., van Hout, M. C., van Steenhoven, A. A., Veldpaus, F. E., and
Janssen, J. D., 1983, ‘‘The Mechanical Properties of Porcine Aortic Valve
Tissues,’’ J. Biomech., 16共5兲, pp. 327–337.
关12兴 Woo, S. L.-Y., Simon, B. R., Kuei, S. C., and Akeson, W. H., 1980, ‘‘Quasi-
Linear Viscoelastic Properties of Normal Articular Cartilage,’’ J. Biomech.
Eng., 102共2兲, pp. 85–90.
关13兴 Thomopoulos, S., Williams, G. R., Gimbel, J. A., Favata, M., and Soslowsky,
L. J., 2003, ‘‘Variation of Biomechanical, Structural, and Compositional Prop-
erties Along the Tendon to Bone Insertion Site,’’ J. Orthop. Res., 21共3兲, pp.
413– 419.
关14兴 Funk, J. R., Hall, G. W., Crandall, J. R., and Pilkey, W. D., 2000, ‘‘Linear and
Quasi-Linear Viscoelastic Characterization of Ankle Ligaments,’’ J. Biomech.
Eng., 122共1兲, pp. 15–22.
关15兴 Toms, S. R., Dakin, G. J., Lemons, J. E., and Eberhardt, A. W., 2002, ‘‘Quasi-
Linear Viscoelastic Behavior of the Human Periodontal Ligament,’’ J. Bio-
mech., 35共10兲, pp. 1411–1415.
Journal of Biomechanical Engineering
View publication stats
关16兴 Thomopoulos, S., Williams, G. R., and Soslowsky, L. J., 2003, ‘‘Tendon to
Bone Healing: Differences in Biomechanical, Structural, and Compositional
Properties Due to a Range of Activity Levels,’’ J. Biomech. Eng., 125共1兲, pp.
106 –113.
关17兴 Elliott, D. M., Robinson, P. S., Gimbel, J. A., Sarver, J. J., Abboud, J. A. et al.,
2003, ‘‘Effect of Altered Matrix Proteins on Quasilinear Viscoelastic Proper-
ties in Transgenic Mouse Tail Tendons,’’ Ann. Biomed. Eng., 31共5兲, pp. 599–
605.
关18兴 Woo, S. L.-Y., Gomez, M. A., and Akeson, W. H., 1981, ‘‘The Time and
History-Dependent Viscoelastic Properties of the Canine Medical Collateral
Ligament,’’ J. Biomech. Eng., 103共4兲, pp. 293–298.
关19兴 Kwan, M. K., Lin, T. H., and Woo, S. L.-Y., 1993, ‘‘On the Viscoelastic
Properties of the Anteromedial Bundle of the Anterior Cruciate Ligament,’’ J.
Biomech., 26共4 –5兲, pp. 447– 452.
关20兴 Johnson, G. A., Tramaglini, D. M., Levine, R. E., Ohno, K., Choi, N. Y., and
Woo, S. L.-Y., 1994, ‘‘Tensile and Viscoelastic Properties of Human Patellar
Tendon,’’ J. Orthop. Res., 12共6兲, pp. 796 – 803.
关21兴 Lin, H. C., Kwan, M. K., and Woo, S. L.-Y., 1987, ‘‘On the Stress Relaxation
Properties of Anterior Cruciate Ligament 共ACL兲,’’ Proceedings, ASME Adv
Bioeng., pp. 5– 6.
关22兴 Nigul, I., and Nigul, U., 1987, ‘‘On Algorithms of Evaluation of Fung’s Re-
laxation Function Parameters,’’ J. Biomech., 20共4兲, pp. 343–352.
关23兴 Abramowitch, S. D., Yagi, M., Tsuda, E., and Woo, S. L.-Y., 2003, ‘‘The
Healing Medial Collateral Ligament Following a Combined Anterior Cruciate
and Medial Collateral Ligament Injury—A Biomechanical Study in a Goat
Model,’’ J. Orthop. Res., 21共6兲, pp. 1124 –1130.
关24兴 Scheffler, S. U., Clineff, T. D., Papageorgiou, C. D., Debski, R. E., Benjamin,
C., and Woo, S. L.-Y., 2001, ‘‘Structure and Function of the Healing Medial
Collateral Ligament in a Goat Model,’’ Ann. Biomed. Eng., 29共2兲, pp. 173–
180.
关25兴 Lee, T. Q., and Woo, S. L.-Y., 1988, ‘‘A New Method for Determining Cross-
Sectional Shape and Area of Soft Tissues,’’ J. Biomech. Eng., 110共2兲, pp.
110–114.
关26兴 Woo, S. L.-Y., Gomez, M. A., Seguchi, Y., Endo, C. M., and Akeson, W. H.,
1983, ‘‘Measurement of Mechanical Properties of Ligament Substance From a
Bone-Ligament-Bone Preparation,’’ J. Orthop. Res., 1共1兲, pp. 22–29.
关27兴 Woo, S. L.-Y., Orlando, C. A., Camp, J. F., and Akeson, W. H., 1986, ‘‘Effects
of Postmortem Storage by Freezing on Ligament Tensile Behavior,’’ J. Bio-
mech., 19共5兲, pp. 399– 404.
关28兴 Woo, S. L.-Y., Danto, M. I., Ohland, K. J., Lee, T. Q., and Newton, P. O., 1990,
‘‘The use of a Laser Micrometer System to Determine the Cross-Sectional
Shape and Area of Ligaments: A Comparative Study With Two Existing Meth-
ods,’’ J. Biomech. Eng., 112共4兲, pp. 426 – 431.
关29兴 Woo, S. L.-Y., 1982, ‘‘Mechanical Properties of Tendons and Ligaments. I.
Quasi-Static and Nonlinear Viscoelastic Properties,’’ Biorheology, 19共3兲, pp.
385–396.
关30兴 Press, W. H., Teukolsky, S. A., Vetterling, W. T., and Flannery, B. P., 1992,
‘‘Chapter 15: Modeling of Data,’’ in Numerical Recipies in C: The Art of
Scientific Computing, Cambridge University Press, New York, NY, pp. 681–
688.
关31兴 Woo, S. L.-Y., Peterson, R. H., Ohland, K. J., Sites, T. J., and Danto, M. I.,
1990, ‘‘The Effects of Strain Rate on the Properties of the Medial Collateral
Ligament in Skeletally Immature and Mature Rabbits: A Biomechanical and
Histological Study,’’ J. Orthop. Res., 8共5兲, pp. 712–721.
关32兴 Yin, F. C., Chew, P. H., and Zeger, S. L., 1986, ‘‘An Approach to Quantifica-
tion of Biaxial Tissue Stress-Strain Data,’’ J. Biomech., 19共1兲, pp. 27–37.
关33兴 Dortmans, L. J., Sauren, A. A., and Rousseau, E. P., 1984, ‘‘Parameter Esti-
mation Using the Quasi-Linear Viscoelastic Model Proposed by Fung,’’ J.
Biomech. Eng., 106共3兲, pp. 198 –203.
关34兴 Sauren, A. A., and Rousseau, E. P., 1983, ‘‘A Concise Sensitivity Analysis of
the Quasi-Linear Viscoelastic Model Proposed by Fung,’’ J. Biomech. Eng.,
105共1兲, pp. 92–95.
关35兴 Lyon, R. M., Lin, H. C., Kwan, M. K., Hollis, J. M., Akeson, W. H., and Woo,
S. L.-Y., 1988, ‘‘Stress Relaxation of the Anterior Cruciate Ligament 共ACL兲
and the Patellar Tendon,’’ Proceedings, 34th Annual Meeting, Orthopaedic
Research Society, Atlanta, GA, p. 81.
FEBRUARY 2004, Vol. 126 Õ 97